fs/overlayfs/copy_up.c

   1 /*
   2  *
   3  * Copyright (C) 2011 Novell Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 as published by
   7  * the Free Software Foundation.
   8  */
   9
  10 #include <linux/module.h>
  11 #include <linux/fs.h>
  12 #include <linux/slab.h>
  13 #include <linux/file.h>
  14 #include <linux/splice.h>
  15 #include <linux/xattr.h>
  16 #include <linux/security.h>
  17 #include <linux/uaccess.h>
  18 #include <linux/sched/signal.h>
  19 #include <linux/cred.h>
  20 #include <linux/namei.h>
  21 #include <linux/fdtable.h>
  22 #include <linux/ratelimit.h>
  23 #include <linux/exportfs.h>
  24 #include "overlayfs.h"
  25 #include "ovl_entry.h"
  26
  27 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
  28
  29 static bool __read_mostly ovl_check_copy_up;
  30 module_param_named(check_copy_up, ovl_check_copy_up, bool,
  31                    S_IWUSR | S_IRUGO);
  32 MODULE_PARM_DESC(ovl_check_copy_up,
  33                  "Warn on copy-up when causing process also has a R/O fd open");
  34
  35 static int ovl_check_fd(const void *data, struct file *f, unsigned int fd)
  36 {
  37         const struct dentry *dentry = data;
  38
  39         if (file_inode(f) == d_inode(dentry))
  40                 pr_warn_ratelimited("overlayfs: Warning: Copying up %pD, but open R/O on fd %u which will cease to be coherent [pid=%d %s]\n",
  41                                     f, fd, current->pid, current->comm);
  42         return 0;
  43 }
  44
  45 /*
  46  * Check the fds open by this process and warn if something like the following
  47  * scenario is about to occur:
  48  *
  49  *      fd1 = open("foo", O_RDONLY);
  50  *      fd2 = open("foo", O_RDWR);
  51  */
  52 static void ovl_do_check_copy_up(struct dentry *dentry)
  53 {
  54         if (ovl_check_copy_up)
  55                 iterate_fd(current->files, 0, ovl_check_fd, dentry);
  56 }
  57
  58 int ovl_copy_xattr(struct dentry *old, struct dentry *new)
  59 {
  60         ssize_t list_size, size, value_size = 0;
  61         char *buf, *name, *value = NULL;
  62         int uninitialized_var(error);
  63         size_t slen;
  64
  65         if (!(old->d_inode->i_opflags & IOP_XATTR) ||
  66             !(new->d_inode->i_opflags & IOP_XATTR))
  67                 return 0;
  68
  69         list_size = vfs_listxattr(old, NULL, 0);
  70         if (list_size <= 0) {
  71                 if (list_size == -EOPNOTSUPP)
  72                         return 0;
  73                 return list_size;
  74         }
  75
  76         buf = kzalloc(list_size, GFP_KERNEL);
  77         if (!buf)
  78                 return -ENOMEM;
  79
  80         list_size = vfs_listxattr(old, buf, list_size);
  81         if (list_size <= 0) {
  82                 error = list_size;
  83                 goto out;
  84         }
  85
  86         for (name = buf; list_size; name += slen) {
  87                 slen = strnlen(name, list_size) + 1;
  88
  89                 /* underlying fs providing us with an broken xattr list? */
  90                 if (WARN_ON(slen > list_size)) {
  91                         error = -EIO;
  92                         break;
  93                 }
  94                 list_size -= slen;
  95
  96                 if (ovl_is_private_xattr(name))
  97                         continue;
  98 retry:
  99                 size = vfs_getxattr(old, name, value, value_size);
 100                 if (size == -ERANGE)
 101                         size = vfs_getxattr(old, name, NULL, 0);
 102
 103                 if (size < 0) {
 104                         error = size;
 105                         break;
 106                 }
 107
 108                 if (size > value_size) {
 109                         void *new;
 110
 111                         new = krealloc(value, size, GFP_KERNEL);
 112                         if (!new) {
 113                                 error = -ENOMEM;
 114                                 break;
 115                         }
 116                         value = new;
 117                         value_size = size;
 118                         goto retry;
 119                 }
 120
 121                 error = security_inode_copy_up_xattr(name);
 122                 if (error < 0 && error != -EOPNOTSUPP)
 123                         break;
 124                 if (error == 1) {
 125                         error = 0;
 126                         continue; /* Discard */
 127                 }
 128                 error = vfs_setxattr(new, name, value, size, 0);
 129                 if (error)
 130                         break;
 131         }
 132         kfree(value);
 133 out:
 134         kfree(buf);
 135         return error;
 136 }
 137
 138 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
 139 {
 140         struct file *old_file;
 141         struct file *new_file;
 142         loff_t old_pos = 0;
 143         loff_t new_pos = 0;
 144         int error = 0;
 145
 146         if (len == 0)
 147                 return 0;
 148
 149         old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
 150         if (IS_ERR(old_file))
 151                 return PTR_ERR(old_file);
 152
 153         new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
 154         if (IS_ERR(new_file)) {
 155                 error = PTR_ERR(new_file);
 156                 goto out_fput;
 157         }
 158
 159         /* Try to use clone_file_range to clone up within the same fs */
 160         error = vfs_clone_file_range(old_file, 0, new_file, 0, len);
 161         if (!error)
 162                 goto out;
 163         /* Couldn't clone, so now we try to copy the data */
 164         error = 0;
 165
 166         /* FIXME: copy up sparse files efficiently */
 167         while (len) {
 168                 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
 169                 long bytes;
 170
 171                 if (len < this_len)
 172                         this_len = len;
 173
 174                 if (signal_pending_state(TASK_KILLABLE, current)) {
 175                         error = -EINTR;
 176                         break;
 177                 }
 178
 179                 bytes = do_splice_direct(old_file, &old_pos,
 180                                          new_file, &new_pos,
 181                                          this_len, SPLICE_F_MOVE);
 182                 if (bytes <= 0) {
 183                         error = bytes;
 184                         break;
 185                 }
 186                 WARN_ON(old_pos != new_pos);
 187
 188                 len -= bytes;
 189         }
 190 out:
 191         if (!error)
 192                 error = vfs_fsync(new_file, 0);
 193         fput(new_file);
 194 out_fput:
 195         fput(old_file);
 196         return error;
 197 }
 198
 199 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
 200 {
 201         struct iattr attr = {
 202                 .ia_valid =
 203                      ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
 204                 .ia_atime = stat->atime,
 205                 .ia_mtime = stat->mtime,
 206         };
 207
 208         return notify_change(upperdentry, &attr, NULL);
 209 }
 210
 211 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
 212 {
 213         int err = 0;
 214
 215         if (!S_ISLNK(stat->mode)) {
 216                 struct iattr attr = {
 217                         .ia_valid = ATTR_MODE,
 218                         .ia_mode = stat->mode,
 219                 };
 220                 err = notify_change(upperdentry, &attr, NULL);
 221         }
 222         if (!err) {
 223                 struct iattr attr = {
 224                         .ia_valid = ATTR_UID | ATTR_GID,
 225                         .ia_uid = stat->uid,
 226                         .ia_gid = stat->gid,
 227                 };
 228                 err = notify_change(upperdentry, &attr, NULL);
 229         }
 230         if (!err)
 231                 ovl_set_timestamps(upperdentry, stat);
 232
 233         return err;
 234 }
 235
 236 static struct ovl_fh *ovl_encode_fh(struct dentry *lower, uuid_be *uuid)
 237 {
 238         struct ovl_fh *fh;
 239         int fh_type, fh_len, dwords;
 240         void *buf;
 241         int buflen = MAX_HANDLE_SZ;
 242
 243         buf = kmalloc(buflen, GFP_TEMPORARY);
 244         if (!buf)
 245                 return ERR_PTR(-ENOMEM);
 246
 247         /*
 248          * We encode a non-connectable file handle for non-dir, because we
 249          * only need to find the lower inode number and we don't want to pay
 250          * the price or reconnecting the dentry.
 251          */
 252         dwords = buflen >> 2;
 253         fh_type = exportfs_encode_fh(lower, buf, &dwords, 0);
 254         buflen = (dwords << 2);
 255
 256         fh = ERR_PTR(-EIO);
 257         if (WARN_ON(fh_type < 0) ||
 258             WARN_ON(buflen > MAX_HANDLE_SZ) ||
 259             WARN_ON(fh_type == FILEID_INVALID))
 260                 goto out;
 261
 262         BUILD_BUG_ON(MAX_HANDLE_SZ + offsetof(struct ovl_fh, fid) > 255);
 263         fh_len = offsetof(struct ovl_fh, fid) + buflen;
 264         fh = kmalloc(fh_len, GFP_KERNEL);
 265         if (!fh) {
 266                 fh = ERR_PTR(-ENOMEM);
 267                 goto out;
 268         }
 269
 270         fh->version = OVL_FH_VERSION;
 271         fh->magic = OVL_FH_MAGIC;
 272         fh->type = fh_type;
 273         fh->flags = OVL_FH_FLAG_CPU_ENDIAN;
 274         fh->len = fh_len;
 275         fh->uuid = *uuid;
 276         memcpy(fh->fid, buf, buflen);
 277
 278 out:
 279         kfree(buf);
 280         return fh;
 281 }
 282
 283 static int ovl_set_origin(struct dentry *dentry, struct dentry *lower,
 284                           struct dentry *upper)
 285 {
 286         struct super_block *sb = lower->d_sb;
 287         uuid_be *uuid = (uuid_be *) &sb->s_uuid;
 288         const struct ovl_fh *fh = NULL;
 289         int err;
 290
 291         /*
 292          * When lower layer doesn't support export operations store a 'null' fh,
 293          * so we can use the overlay.origin xattr to distignuish between a copy
 294          * up and a pure upper inode.
 295          */
 296         if (sb->s_export_op && sb->s_export_op->fh_to_dentry &&
 297             uuid_be_cmp(*uuid, NULL_UUID_BE)) {
 298                 fh = ovl_encode_fh(lower, uuid);
 299                 if (IS_ERR(fh))
 300                         return PTR_ERR(fh);
 301         }
 302
 303         err = ovl_do_setxattr(upper, OVL_XATTR_ORIGIN, fh, fh ? fh->len : 0, 0);
 304         kfree(fh);
 305
 306         return err;
 307 }
 308
 309 static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
 310                               struct dentry *dentry, struct path *lowerpath,
 311                               struct kstat *stat, const char *link,
 312                               struct kstat *pstat, bool tmpfile)
 313 {
 314         struct inode *wdir = workdir->d_inode;
 315         struct inode *udir = upperdir->d_inode;
 316         struct dentry *newdentry = NULL;
 317         struct dentry *upper = NULL;
 318         struct dentry *temp = NULL;
 319         int err;
 320         const struct cred *old_creds = NULL;
 321         struct cred *new_creds = NULL;
 322         struct cattr cattr = {
 323                 /* Can't properly set mode on creation because of the umask */
 324                 .mode = stat->mode & S_IFMT,
 325                 .rdev = stat->rdev,
 326                 .link = link
 327         };
 328
 329         upper = lookup_one_len(dentry->d_name.name, upperdir,
 330                                dentry->d_name.len);
 331         err = PTR_ERR(upper);
 332         if (IS_ERR(upper))
 333                 goto out;
 334
 335         err = security_inode_copy_up(dentry, &new_creds);
 336         if (err < 0)
 337                 goto out1;
 338
 339         if (new_creds)
 340                 old_creds = override_creds(new_creds);
 341
 342         if (tmpfile)
 343                 temp = ovl_do_tmpfile(upperdir, stat->mode);
 344         else
 345                 temp = ovl_lookup_temp(workdir, dentry);
 346         err = PTR_ERR(temp);
 347         if (IS_ERR(temp))
 348                 goto out1;
 349
 350         err = 0;
 351         if (!tmpfile)
 352                 err = ovl_create_real(wdir, temp, &cattr, NULL, true);
 353
 354         if (new_creds) {
 355                 revert_creds(old_creds);
 356                 put_cred(new_creds);
 357         }
 358
 359         if (err)
 360                 goto out2;
 361
 362         if (S_ISREG(stat->mode)) {
 363                 struct path upperpath;
 364
 365                 ovl_path_upper(dentry, &upperpath);
 366                 BUG_ON(upperpath.dentry != NULL);
 367                 upperpath.dentry = temp;
 368
 369                 if (tmpfile) {
 370                         inode_unlock(udir);
 371                         err = ovl_copy_up_data(lowerpath, &upperpath,
 372                                                stat->size);
 373                         inode_lock_nested(udir, I_MUTEX_PARENT);
 374                 } else {
 375                         err = ovl_copy_up_data(lowerpath, &upperpath,
 376                                                stat->size);
 377                 }
 378
 379                 if (err)
 380                         goto out_cleanup;
 381         }
 382
 383         err = ovl_copy_xattr(lowerpath->dentry, temp);
 384         if (err)
 385                 goto out_cleanup;
 386
 387         inode_lock(temp->d_inode);
 388         err = ovl_set_attr(temp, stat);
 389         inode_unlock(temp->d_inode);
 390         if (err)
 391                 goto out_cleanup;
 392
 393         /*
 394          * Store identifier of lower inode in upper inode xattr to
 395          * allow lookup of the copy up origin inode.
 396          */
 397         err = ovl_set_origin(dentry, lowerpath->dentry, temp);
 398         if (err)
 399                 goto out_cleanup;
 400
 401         if (tmpfile)
 402                 err = ovl_do_link(temp, udir, upper, true);
 403         else
 404                 err = ovl_do_rename(wdir, temp, udir, upper, 0);
 405         if (err)
 406                 goto out_cleanup;
 407
 408         newdentry = dget(tmpfile ? upper : temp);
 409         ovl_dentry_update(dentry, newdentry);
 410         ovl_inode_update(d_inode(dentry), d_inode(newdentry));
 411
 412         /* Restore timestamps on parent (best effort) */
 413         ovl_set_timestamps(upperdir, pstat);
 414 out2:
 415         dput(temp);
 416 out1:
 417         dput(upper);
 418 out:
 419         return err;
 420
 421 out_cleanup:
 422         if (!tmpfile)
 423                 ovl_cleanup(wdir, temp);
 424         goto out2;
 425 }
 426
 427 /*
 428  * Copy up a single dentry
 429  *
 430  * All renames start with copy up of source if necessary.  The actual
 431  * rename will only proceed once the copy up was successful.  Copy up uses
 432  * upper parent i_mutex for exclusion.  Since rename can change d_parent it
 433  * is possible that the copy up will lock the old parent.  At that point
 434  * the file will have already been copied up anyway.
 435  */
 436 static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 437                            struct path *lowerpath, struct kstat *stat)
 438 {
 439         DEFINE_DELAYED_CALL(done);
 440         struct dentry *workdir = ovl_workdir(dentry);
 441         int err;
 442         struct kstat pstat;
 443         struct path parentpath;
 444         struct dentry *lowerdentry = lowerpath->dentry;
 445         struct dentry *upperdir;
 446         const char *link = NULL;
 447         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 448
 449         if (WARN_ON(!workdir))
 450                 return -EROFS;
 451
 452         ovl_do_check_copy_up(lowerdentry);
 453
 454         ovl_path_upper(parent, &parentpath);
 455         upperdir = parentpath.dentry;
 456
 457         err = vfs_getattr(&parentpath, &pstat,
 458                           STATX_ATIME | STATX_MTIME, AT_STATX_SYNC_AS_STAT);
 459         if (err)
 460                 return err;
 461
 462         if (S_ISLNK(stat->mode)) {
 463                 link = vfs_get_link(lowerdentry, &done);
 464                 if (IS_ERR(link))
 465                         return PTR_ERR(link);
 466         }
 467
 468         /* Should we copyup with O_TMPFILE or with workdir? */
 469         if (S_ISREG(stat->mode) && ofs->tmpfile) {
 470                 err = ovl_copy_up_start(dentry);
 471                 /* err < 0: interrupted, err > 0: raced with another copy-up */
 472                 if (unlikely(err)) {
 473                         pr_debug("ovl_copy_up_start(%pd2) = %i\n", dentry, err);
 474                         if (err > 0)
 475                                 err = 0;
 476                         goto out_done;
 477                 }
 478
 479                 inode_lock_nested(upperdir->d_inode, I_MUTEX_PARENT);
 480                 err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
 481                                          stat, link, &pstat, true);
 482                 inode_unlock(upperdir->d_inode);
 483                 ovl_copy_up_end(dentry);
 484                 goto out_done;
 485         }
 486
 487         err = -EIO;
 488         if (lock_rename(workdir, upperdir) != NULL) {
 489                 pr_err("overlayfs: failed to lock workdir+upperdir\n");
 490                 goto out_unlock;
 491         }
 492         if (ovl_dentry_upper(dentry)) {
 493                 /* Raced with another copy-up?  Nothing to do, then... */
 494                 err = 0;
 495                 goto out_unlock;
 496         }
 497
 498         err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
 499                                  stat, link, &pstat, false);
 500 out_unlock:
 501         unlock_rename(workdir, upperdir);
 502 out_done:
 503         do_delayed_call(&done);
 504
 505         return err;
 506 }
 507
 508 int ovl_copy_up_flags(struct dentry *dentry, int flags)
 509 {
 510         int err = 0;
 511         const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
 512
 513         while (!err) {
 514                 struct dentry *next;
 515                 struct dentry *parent;
 516                 struct path lowerpath;
 517                 struct kstat stat;
 518                 enum ovl_path_type type = ovl_path_type(dentry);
 519
 520                 if (OVL_TYPE_UPPER(type))
 521                         break;
 522
 523                 next = dget(dentry);
 524                 /* find the topmost dentry not yet copied up */
 525                 for (;;) {
 526                         parent = dget_parent(next);
 527
 528                         type = ovl_path_type(parent);
 529                         if (OVL_TYPE_UPPER(type))
 530                                 break;
 531
 532                         dput(next);
 533                         next = parent;
 534                 }
 535
 536                 ovl_path_lower(next, &lowerpath);
 537                 err = vfs_getattr(&lowerpath, &stat,
 538                                   STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
 539                 /* maybe truncate regular file. this has no effect on dirs */
 540                 if (flags & O_TRUNC)
 541                         stat.size = 0;
 542                 if (!err)
 543                         err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
 544
 545                 dput(parent);
 546                 dput(next);
 547         }
 548         revert_creds(old_cred);
 549
 550         return err;
 551 }
 552
 553 int ovl_copy_up(struct dentry *dentry)
 554 {
 555         return ovl_copy_up_flags(dentry, 0);
 556 }